(152j) Machine-Learning Enabled Screening of MOFs for Ion Selective Membranes

The design of ion selective materials with improved separations efficacy and efficiency is critical for the development of novel membrane technologies in water purification and desalination. Metal-organic frameworks (MOFs) serve as promising materials for ion separation membranes given their rigid structure and high tunability, allowing for tailored design of pore sizes, geometries, and chemical topology. Inspired by the ultra-high selectivity exhibited in biological ion channels, we aim to understand the fundamental design rules in material structure and chemical patterning which leads to these biomimetic functionalities. Towards these goals, we screen the Computationally-Ready Experimental (CoRE) MOF database of 9,598 experimentally known and synthesizable MOFs to find a set of promising materials for membranes based on structural and chemical characteristics, water stability, and ion transport. In order to determine water stability of MOF candidates, we undertake a transfer learning approach to expand existing machine-learning MOF thermal stability and solvent removal stability models to water stability. Ion-MOF binding energies calculated from DFT and ion transport properties obtained from classical molecular dynamics simulations provide further insight towards the underlying driving forces for ion selectivity and transport. Ultimately, a curated set of promising MOF candidates for ion selective membranes is identified for deeper experimental and computational investigation.